public void EmbedIDRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 30);
int outputCount = Mother.Dice.Next(1, 30);
int batchCount = Mother.Dice.Next(1, 5);
int[,] input = (int[, ])Enumerable.Repeat(0, batchCount * inputCount).ToNdArray(batchCount, inputCount);
input[0, 0] = 1;
Real[,,] dummyGy = (Real[, , ])Initializer.GetRealNdArray(new[] { batchCount, inputCount, outputCount });
Real[,] w = (Real[, ])Initializer.GetRealNdArray(new[] { inputCount, outputCount });
//Chainer
NChainer.EmbedID <Real> cEmbedId = new NChainer.EmbedID <Real>(inputCount, outputCount, Real.ToBaseNdArray(w));
Variable <int> cX = new Variable <int>(input);
Variable <Real> cY = cEmbedId.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.EmbedID embedId = new KelpNet.EmbedID(inputCount, outputCount, w);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { inputCount }, batchCount);
NdArray y = embedId.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, , ])cY.Data);
Real[] cWgrad = Real.ToRealArray((Real[, ])cEmbedId.W.Grad);
//許容範囲を算出
double delta = 0.00001;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//W.grad
Assert.AreEqual(cWgrad.Length, embedId.Weight.Grad.Length);
for (int i = 0; i < embedId.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], embedId.Weight.Grad[i], delta);
}
}
public void EmbedIDRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 30);
int outputCount = Mother.Dice.Next(1, 30);
int batchCount = Mother.Dice.Next(1, 5);
int[,] input = (int[, ])Enumerable.Repeat(0, batchCount * inputCount).ToNdArray(batchCount, inputCount);
input[0, 0] = 1;
Real[,,] dummyGy = Initializer.GetRandomValues <Real[, , ]>(batchCount, inputCount, outputCount);
Real[,] w = Initializer.GetRandomValues <Real[, ]>(inputCount, outputCount);
//Chainer
NChainer.EmbedID <Real> cEmbedId = new NChainer.EmbedID <Real>(inputCount, outputCount, w);
Variable <int> cX = new Variable <int>(input);
Variable <Real> cY = cEmbedId.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
EmbedID <Real> embedId = new EmbedID <Real>(inputCount, outputCount, w);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = embedId.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, , ])cY.Data).Flatten();
Real[] cWgrad = ((Real[, ])cEmbedId.W.Grad).Flatten();
//許容範囲を算出
Real delta = 0.00001f;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//W.grad
Assert.AreEqual(cWgrad.Length, embedId.Weight.Grad.Length);
for (int i = 0; i < embedId.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], embedId.Weight.Grad[i], delta);
}
}
public void SoftmaxCrossEntropyRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int channel = Mother.Dice.Next(2, 5);
int width = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
Real[,,,] inputA = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, channel, height, width });
int[,,] inputB = (int[, , ])Enumerable.Repeat(0, batchCount * height * width).ToNdArray(batchCount, height, width);
for (int b = 0; b < batchCount; b++)
{
inputB[b, Mother.Dice.Next(height), Mother.Dice.Next(width)] = 1;
}
//chainer
NChainer.SoftmaxCrossEntropy <Real> cSoftmaxCrossEntropy = new NChainer.SoftmaxCrossEntropy <Real>();
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(inputA));
Variable <int> cY = new Variable <int>(inputB);
Variable <Real> cZ = cSoftmaxCrossEntropy.Forward(cX, cY);
cZ.Backward();
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad);
//KelpNet
KelpNet.SoftmaxCrossEntropy softmaxCrossEntropy = new KelpNet.SoftmaxCrossEntropy();
NdArray x = new NdArray(Real.ToRealArray(inputA), new[] { channel, height, width }, batchCount);
NdArray y = new NdArray(Real.ToRealArray(inputB), new[] { height, width }, batchCount);
//KelpNetはBackwaward側のみEvaluateで実行される
NdArray z = softmaxCrossEntropy.Evaluate(x, y);
//許容範囲を算出(内部の割引順が違うため誤差が大きい)
double delta = 0.00001;
//Loss
Assert.AreEqual(cZ.Data[0], z.Data[0], delta);
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void ConcatRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int widthA = Mother.Dice.Next(1, 16);
int widthB = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
int axis = 3;
Real[,,,] inputA = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthA);
Real[,,,] inputB = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthB);
Real[,,,] dummyGy = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthA + widthB);
//chainer
NChainer.Concat <Real> cConcat = new NChainer.Concat <Real>(axis);
Variable <Real> cX = new Variable <Real>(inputA);
Variable <Real> cY = new Variable <Real>(inputB);
Variable <Real> cZ = cConcat.Forward(cX, cY);
cZ.Grad = dummyGy;
cZ.Backward();
//KelpNet
Concat <Real> concat = new Concat <Real>(axis - 1);//Chainerと異なり1次元目を暗黙的にBatchとみなさないため
NdArray <Real> x = new NdArray <Real>(inputA, asBatch: true);
NdArray <Real> y = new NdArray <Real>(inputB, asBatch: true);
NdArray <Real> z = concat.Forward(x, y)[0];
z.Grad = dummyGy.Flatten();
z.Backward();
Real[] cZdata = ((Real[, , , ])cZ.Data).Flatten();
//Copyが必要
Real[] cXgrad = ((Real[, , , ])cX.Grad.Copy()).Flatten();
Real[] cYgrad = ((Real[, , , ])cY.Grad.Copy()).Flatten();
//許容範囲を算出
Real delta = 0.00001f;
//z
Assert.AreEqual(cZdata.Length, z.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cZdata[i], z.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//y.grad
Assert.AreEqual(cYgrad.Length, y.Grad.Length);
for (int i = 0; i < y.Grad.Length; i++)
{
Assert.AreEqual(cYgrad[i], y.Grad[i], delta);
}
}
public void LSTMRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int inputCount = Mother.Dice.Next(1, 5);
int outputCount = Mother.Dice.Next(1, 5);
Real[,] input = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, inputCount });
Real[,] dummyGy = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, outputCount });
Real[,] upwardInit = (Real[, ])Initializer.GetRealNdArray(new[] { outputCount, inputCount });
Real[,] lateralInit = (Real[, ])Initializer.GetRealNdArray(new[] { outputCount, outputCount });
Real[,,] biasInit = (Real[, , ])Initializer.GetRealNdArray(new[] { 1, outputCount, 1 });
Real[,,] forgetBiasInit = (Real[, , ])Initializer.GetRealNdArray(new[] { 1, outputCount, 1 });
//Chainer
NChainer.LSTM <Real> clstm = new NChainer.LSTM <Real>(inputCount, outputCount, Real.ToBaseNdArray(lateralInit), Real.ToBaseNdArray(upwardInit), Real.ToBaseNdArray(biasInit), Real.ToBaseNdArray(forgetBiasInit));
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = clstm.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.LSTM lstm = new KelpNet.LSTM(inputCount, outputCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { inputCount }, batchCount);
NdArray y = lstm.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
//許容範囲を算出
double delta = 0.00001;
Real[] cYdata = Real.ToRealArray((Real[, ])cY.Data);
Real[] cXgrad = Real.ToRealArray((Real[, ])cX.Grad);
Real[] cupwardWGrad = Real.ToRealArray((Real[, ])clstm.upward.W.Grad);
Real[] cupwardbGrad = (Real[])clstm.upward.b.Grad;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < cYdata.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < cXgrad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//W.grad
int wLen = lstm.upward.Weight.Grad.Length;
Assert.AreEqual(cupwardWGrad.Length, lstm.upward.Weight.Grad.Length);
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
int bLen = lstm.upward.Bias.Length;
Assert.AreEqual(cupwardbGrad.Length, lstm.upward.Bias.Grad.Length);
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
//////////////////////////////////////////////////////////////////////////////////////////
// 1度目はlateralに値はない //
//////////////////////////////////////////////////////////////////////////////////////////
//Real[] clateralWGrad = Real.ToRealArray((Real[,])clstm.lateral.W.Grad);
//Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
//for (int i = 0; i < wLen; i++)
//{
// Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
//}
//////////////////////////////////////////////////////////////////////////////////////////
///////////
//２周目 //
///////////
Real[,] input2 = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, inputCount });
Real[,] dummyGy2 = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, outputCount });
//Chainer
Variable <Real> cX2 = new Variable <Real>(Real.ToBaseNdArray(input2));
Variable <Real> cY2 = clstm.Forward(cX2);
cY2.Grad = Real.ToBaseNdArray(dummyGy2);
cY2.Backward();
//KelpNet
NdArray x2 = new NdArray(Real.ToRealArray(input2), new[] { inputCount }, batchCount);
NdArray y2 = lstm.Forward(x2)[0];
y2.Grad = Real.ToRealArray(dummyGy2);
y2.Backward();
Real[] cYdata2 = Real.ToRealArray((Real[, ])cY2.Data);
Real[] cXgrad2 = Real.ToRealArray((Real[, ])cX2.Grad);
Real[] cupwardWGrad2 = Real.ToRealArray((Real[, ])clstm.upward.W.Grad);
Real[] cupwardbGrad2 = (Real[])clstm.upward.b.Grad;
Real[] clateralWGrad = Real.ToRealArray((Real[, ])clstm.lateral.W.Grad);
//y
Assert.AreEqual(cYdata2.Length, y2.Data.Length);
for (int i = 0; i < cYdata2.Length; i++)
{
Assert.AreEqual(cYdata2[i], y2.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad2.Length, x2.Grad.Length);
for (int i = 0; i < cXgrad2.Length; i++)
{
Assert.AreEqual(cXgrad2[i], x2.Grad[i], delta);
}
//W.grad
Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
for (int i = 0; i < clateralWGrad.Length; i++)
{
Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
}
//経由が多いため誤差が大きい
delta = 1.0;
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad2[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad2[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
}
public void AdamRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 50);
int outputCount = Mother.Dice.Next(2, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = (Real[,])Initializer.GetRealNdArray(new[] { batchCount, inputCount });
Real[,] dummyGy = (Real[,])Initializer.GetRealNdArray(new[] { batchCount, outputCount });
Real[,] w = (Real[,])Initializer.GetRealNdArray(new[] { outputCount, inputCount });
Real[] b = Initializer.GetRealArray(outputCount);
float alpha = (float)Mother.Dice.NextDouble(); //0.001f
float beta1 = (float)Mother.Dice.NextDouble(); //0.9f;
float beta2 = (float)Mother.Dice.NextDouble(); //0.999f;
float eps = (float)Mother.Dice.NextDouble(); //1e-08f;
float eta = (float)Mother.Dice.NextDouble(); //1.0f;
//Chainer
NChainer.Linear<Real> cLinear = new NChainer.Linear<Real>(inputCount, outputCount, false, Real.ToBaseNdArray(w), Real.ToBaseArray(b));
NChainer.Adam<Real> cAdam = new NChainer.Adam<Real>(alpha, beta1, beta2, eps, eta);
cAdam.Setup(cLinear);
Variable<Real> cX = new Variable<Real>(Real.ToBaseNdArray(input));
Variable<Real> cY = cLinear.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
cAdam.Update();
//KelpNet
KelpNet.CL.Linear linear = new KelpNet.CL.Linear(inputCount, outputCount, false, w, b);
KelpNet.Adam adam = new Adam(alpha, beta1, beta2, eps, eta);
adam.SetUp(linear);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { inputCount }, batchCount);
NdArray y = linear.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
adam.Update();
Real[] cW = Real.ToRealArray((Real[,])cLinear.W.Data);
Real[] cb = (Real[])cLinear.b.Data;
//許容範囲を算出
double delta = 0.00001;
//W.grad
Assert.AreEqual(cW.Length, linear.Weight.Data.Length);
for (int i = 0; i < linear.Weight.Data.Length; i++)
{
Assert.AreEqual(cW[i], linear.Weight.Data[i], delta);
}
//b.grad
Assert.AreEqual(cb.Length, linear.Bias.Data.Length);
for (int i = 0; i < linear.Bias.Data.Length; i++)
{
Assert.AreEqual(cb[i], linear.Bias.Data[i], delta);
}
}
public void RandomTest(bool gpuEnable)
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int inChCount = Mother.Dice.Next(1, 5);
int outChCount = Mother.Dice.Next(1, 5);
int wideSize = Mother.Dice.Next(8, 32);
int heightSize = Mother.Dice.Next(8, 32);
int kWidth = Mother.Dice.Next(1, 5);
int kHeight = Mother.Dice.Next(1, 5);
int strideX = Mother.Dice.Next(1, 5);
int strideY = Mother.Dice.Next(1, 5);
int padX = Mother.Dice.Next(0, 5);
int padY = Mother.Dice.Next(0, 5);
int outputHeight = (int)Math.Floor((heightSize - kHeight + padY * 2.0) / strideY) + 1;
int outputWidth = (int)Math.Floor((wideSize - kWidth + padX * 2.0) / strideX) + 1;
Real[,,,] input = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, inChCount, heightSize, wideSize });
Real[,,,] dummyGy = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, outChCount, outputHeight, outputWidth });
Real[,,,] w = (Real[, , , ])Initializer.GetRealNdArray(new[] { outChCount, inChCount, kHeight, kWidth });
Real[] b = Initializer.GetRealArray(outChCount);
//Chainer
NChainer.Convolution2D <Real> cConvolotion2D = new NChainer.Convolution2D <Real>(
inChCount, outChCount,
new[] { kHeight, kWidth },
new[] { strideY, strideX },
new[] { padY, padX },
false,
Real.ToBaseNdArray(w),
Real.ToBaseArray(b));
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cConvolotion2D.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.CL.Convolution2D convolution2D = new KelpNet.CL.Convolution2D(
inChCount, outChCount,
new[] { kWidth, kHeight },
new[] { strideX, strideY },
new[] { padX, padY },
false, w, b, gpuEnable: gpuEnable);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { inChCount, heightSize, wideSize }, batchCount);
NdArray y = convolution2D.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, , , ])cY.Data.Copy());
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad.Copy());
Real[] cWgrad = Real.ToRealArray((Real[, , , ])cConvolotion2D.W.Grad);
Real[] cbgrad = (Real[])cConvolotion2D.b.Grad;
//許容範囲を算出
double delta = 0.00001;
Assert.AreEqual(cYdata.Length, y.Data.Length);
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
Assert.AreEqual(cWgrad.Length, convolution2D.Weight.Grad.Length);
Assert.AreEqual(cbgrad.Length, convolution2D.Bias.Grad.Length);
//y
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
delta = 0.1;
//W.grad
for (int i = 0; i < convolution2D.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], convolution2D.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < convolution2D.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], convolution2D.Bias.Grad[i], delta);
}
}
public void RandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = 1 + Mother.Dice.Next() % 49;
int outputCount = 1 + Mother.Dice.Next() % 49;
Real[,] input = (Real[, ])Initializer.GetRealNdArray(new[] { 1, inputCount });
Real[,] dummyGy = (Real[, ])Initializer.GetRealNdArray(new[] { 1, outputCount });
Real[,] w = (Real[, ])Initializer.GetRealNdArray(new[] { outputCount, inputCount });
Real[] b = Initializer.GetRealArray(outputCount);
//Chainer
NChainer.Linear <Real> cLinear = new NChainer.Linear <Real>(inputCount, outputCount, false, Real.ToBaseNdArray(w), Real.ToBaseArray(b));
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cLinear.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.Linear linear = new KelpNet.Linear(inputCount, outputCount, false, w, b);
NdArray x = new NdArray(input);
NdArray y = linear.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, ])cY.Data);
Real[] cXgrad = Real.ToRealArray((Real[, ])cX.Grad);
Real[] cWgrad = Real.ToRealArray((Real[, ])cLinear.W.Grad);
Real[] cbgrad = (Real[])cLinear.b.Grad;
//許容範囲を算出
double delta = 0.00001;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//W.grad
Assert.AreEqual(cWgrad.Length, linear.Weight.Grad.Length);
for (int i = 0; i < linear.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], linear.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad.Length, linear.Bias.Grad.Length);
for (int i = 0; i < linear.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], linear.Bias.Grad[i], delta);
}
}
public void SwishRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int ioCount = Mother.Dice.Next(1, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = (Real[,])Initializer.GetRealNdArray(new[] { batchCount, ioCount });
Real[,] dummyGy = (Real[,])Initializer.GetRealNdArray(new[] { batchCount, ioCount });
Real beta = Mother.Dice.NextDouble();
//Chainer
NChainer.Swish<Real> cSwish = new NChainer.Swish<Real>(new[] { ioCount }, beta);
Variable<Real> cX = new Variable<Real>(Real.ToBaseNdArray(input));
Variable<Real> cY = cSwish.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.Swish swish = new KelpNet.Swish(new[] { ioCount }, beta);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { ioCount }, batchCount);
NdArray y = swish.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[,])cY.Data);
Real[] cXgrad = Real.ToRealArray((Real[,])cX.Grad);
Real[] cbgrad = (Real[])cSwish.beta.Grad;
//許容範囲を算出
double delta = 0.00001;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad.Length, swish.Beta.Grad.Length);
for (int i = 0; i < swish.Beta.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], swish.Beta.Grad[i], delta);
}
}
public void LSTMRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int inputCount = Mother.Dice.Next(1, 5);
int outputCount = Mother.Dice.Next(1, 5);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, inputCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, outputCount);
Real[,] upwardInit = Initializer.GetRandomValues <Real[, ]>(outputCount, inputCount);
Real[,] lateralInit = Initializer.GetRandomValues <Real[, ]>(outputCount, outputCount);
Real[,,] biasInit = Initializer.GetRandomValues <Real[, , ]>(1, outputCount, 1);
Real[,,] forgetBiasInit = Initializer.GetRandomValues <Real[, , ]>(1, outputCount, 1);
//Chainer
NChainer.LSTM <Real> clstm = new NChainer.LSTM <Real>(inputCount, outputCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = clstm.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
LSTM <Real> lstm = new LSTM <Real>(inputCount, outputCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = lstm.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
//許容範囲を算出
Real delta = 0.00001f;
Real[] cYdata = ((Real[, ])cY.Data).Flatten();
Real[] cXgrad = ((Real[, ])cX.Grad).Flatten();
Real[] cupwardWGrad = ((Real[, ])clstm.upward.W.Grad).Flatten();
Real[] cupwardbGrad = (Real[])clstm.upward.b.Grad;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < cYdata.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < cXgrad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//W.grad
int wLen = lstm.upward.Weight.Grad.Length;
Assert.AreEqual(cupwardWGrad.Length, lstm.upward.Weight.Grad.Length);
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
int bLen = lstm.upward.Bias.Length;
Assert.AreEqual(cupwardbGrad.Length, lstm.upward.Bias.Grad.Length);
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
//////////////////////////////////////////////////////////////////////////////////////////
// 1度目はlateralに値はない //
//////////////////////////////////////////////////////////////////////////////////////////
//Real[] clateralWGrad = Real.ToRealArray((Real[,])clstm.lateral.W.Grad);
//Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
//for (int i = 0; i < wLen; i++)
//{
// Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
//}
//////////////////////////////////////////////////////////////////////////////////////////
///////////
//２周目 //
///////////
Real[,] input2 = Initializer.GetRandomValues <Real[, ]>(batchCount, inputCount);
Real[,] dummyGy2 = Initializer.GetRandomValues <Real[, ]>(batchCount, outputCount);
//Chainer
Variable <Real> cX2 = new Variable <Real>(input2);
Variable <Real> cY2 = clstm.Forward(cX2);
cY2.Grad = dummyGy2;
cY2.Backward();
//KelpNet
NdArray <Real> x2 = new NdArray <Real>(input2, asBatch: true);
NdArray <Real> y2 = lstm.Forward(x2)[0];
y2.Grad = dummyGy2.Flatten();
y2.Backward();
Real[] cYdata2 = ((Real[, ])cY2.Data).Flatten();
Real[] cXgrad2 = ((Real[, ])cX2.Grad).Flatten();
Real[] cupwardWGrad2 = ((Real[, ])clstm.upward.W.Grad).Flatten();
Real[] cupwardbGrad2 = (Real[])clstm.upward.b.Grad;
Real[] clateralWGrad = ((Real[, ])clstm.lateral.W.Grad).Flatten();
//y
Assert.AreEqual(cYdata2.Length, y2.Data.Length);
for (int i = 0; i < cYdata2.Length; i++)
{
Assert.AreEqual(cYdata2[i], y2.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad2.Length, x2.Grad.Length);
for (int i = 0; i < cXgrad2.Length; i++)
{
Assert.AreEqual(cXgrad2[i], x2.Grad[i], delta);
}
//W.grad
Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
for (int i = 0; i < clateralWGrad.Length; i++)
{
Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
}
//経由が多いため誤差が大きい
delta = 1.0f;
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad2[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad2[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
}
public void BasicMathRandomTest()
{
Python.Initialize();
Chainer.Initialize();
//Make random value.
Real[] val = { 1 + Mother.Dice.Next() };
//Chainer
Variable <Real> cX = new Variable <Real>(val);
//Add
Variable <Real> cAdd = 2 + cX + cX + 2;
cAdd.Backward();
Real[] pyGadd = (Real[])cX.Grad;
//Mul
Variable <Real> cMul = 2 * cX * cX * 3;
cMul.Backward();
Real[] pyGmul = (Real[])cX.Grad;
//Sub
Variable <Real> cSub = 30 - cX - cX - 2;
cSub.Backward();
Real[] pyGsub = (Real[])cX.Grad;
//Div
Variable <Real> cDiv = 50 / cX / cX / 2;
cDiv.Backward();
Real[] pyGdiv = (Real[])cX.Grad;
//KelpNet
NdArray <Real> x = new NdArray <Real>(val);
//Add
NdArray <Real> add = 2 + x + x + 2;
add.Backward();
Real[] gadd = x.Grad.ToArray(); //このToArrayはコピーのため
//mul
NdArray <Real> mul = 2 * x * x * 3;
mul.Backward();
Real[] gmul = x.Grad.ToArray();
//sub
NdArray <Real> sub = 30 - x - x - 2;
sub.Backward();
Real[] gsub = x.Grad.ToArray();
//mul
NdArray <Real> div = 50 / x / x / 2;
div.Backward();
Real[] gdiv = x.Grad.ToArray();
//Check
CollectionAssert.AreEqual(pyGadd, gadd);
CollectionAssert.AreEqual(pyGmul, gmul);
CollectionAssert.AreEqual(pyGsub, gsub);
CollectionAssert.AreEqual(pyGdiv, gdiv);
}
public void RnnLSTMRandomTest()
{
Python.Initialize();
Chainer.Initialize();
Real[,] input = { { 1.0 }, { 3.0 }, { 5.0 }, { 7.0 }, { 9.0 } };
Real[,] teach = { { 3.0 }, { 5.0 }, { 7.0 }, { 9.0 }, { 11.0 } };
Real[,] input2 = { { 3.0 }, { 5.0 }, { 7.0 }, { 9.0 }, { 11.0 } };
Real[,] teach2 = { { 5.0 }, { 7.0 }, { 9.0 }, { 11.0 }, { 13.0 } };
int outputCount = 1;
int inputCount = 1;
int hiddenCount = 2;
Real[,] upwardInit = (Real[, ])Initializer.GetRealNdArray(new[] { hiddenCount, hiddenCount });
Real[,] lateralInit = (Real[, ])Initializer.GetRealNdArray(new[] { hiddenCount, hiddenCount });
Real[,,] biasInit = (Real[, , ])Initializer.GetRealNdArray(new[] { 1, hiddenCount, 1 });
Real[,,] forgetBiasInit = (Real[, , ])Initializer.GetRealNdArray(new[] { 1, hiddenCount, 1 });
//Chainer
Real[,] w1 = (Real[, ])Initializer.GetRealNdArray(new[] { hiddenCount, inputCount });
Real[] b1 = Initializer.GetRealArray(hiddenCount);
//Chainer
NChainer.Linear <Real> cLinear1 = new NChainer.Linear <Real>(inputCount, hiddenCount, false, Real.ToBaseNdArray(w1), Real.ToBaseArray(b1));
NChainer.LSTM <Real> cLstm = new NChainer.LSTM <Real>(hiddenCount, hiddenCount, Real.ToBaseNdArray(lateralInit), Real.ToBaseNdArray(upwardInit), Real.ToBaseNdArray(biasInit), Real.ToBaseNdArray(forgetBiasInit));
Real[,] w2 = (Real[, ])Initializer.GetRealNdArray(new[] { outputCount, hiddenCount });
Real[] b2 = Initializer.GetRealArray(outputCount);
NChainer.Linear <Real> cLinear2 = new NChainer.Linear <Real>(hiddenCount, outputCount, false, Real.ToBaseNdArray(w2), Real.ToBaseArray(b2));
Variable <Real> cX1 = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY11 = cLinear1.Forward(cX1);
Variable <Real> cY12 = cLstm.Forward(cY11);
Variable <Real> cY13 = cLinear2.Forward(cY12);
Variable <Real> cT = new Variable <Real>(Real.ToBaseNdArray(teach));
Variable <Real> cLoss = new NChainer.MeanSquaredError <Real>().Forward(cY13, cT);
cLoss.Backward();
//KelpNet
KelpNet.CL.Linear linear1 = new KelpNet.CL.Linear(inputCount, hiddenCount, false, w1, b1);
KelpNet.LSTM lstm = new KelpNet.LSTM(hiddenCount, hiddenCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
KelpNet.CL.Linear linear2 = new KelpNet.CL.Linear(hiddenCount, outputCount, false, w2, b2);
NdArray x1 = new NdArray(Real.ToRealArray(input), new[] { 1 }, 5);
NdArray y11 = linear1.Forward(x1)[0];
NdArray y12 = lstm.Forward(y11)[0];
NdArray y13 = linear2.Forward(y12)[0];
NdArray t = new NdArray(Real.ToRealArray(teach), new[] { 1 }, 5);
NdArray loss = new KelpNet.MeanSquaredError().Evaluate(y13, t);
y13.Backward();
Real[] cY11data = Real.ToRealArray((Real[, ])cY11.Data);
Real[] cY12data = Real.ToRealArray((Real[, ])cY12.Data);
Real[] cY13data = Real.ToRealArray((Real[, ])cY13.Data);
Real[] cXgrad = Real.ToRealArray((Real[, ])cX1.Grad);
Real[] cupwardWGrad = Real.ToRealArray((Real[, ])cLstm.upward.W.Grad);
Real[] cupwardbGrad = (Real[])cLstm.upward.b.Grad;
//許容範囲を設定
double delta = 0.00001;
//y11
Assert.AreEqual(cY11data.Length, y11.Data.Length);
for (int i = 0; i < cY11data.Length; i++)
{
Assert.AreEqual(cY11data[i], y11.Data[i], delta);
}
//y12
Assert.AreEqual(cY12data.Length, y12.Data.Length);
for (int i = 0; i < cY12data.Length; i++)
{
Assert.AreEqual(cY12data[i], y12.Data[i], delta);
}
//y13
Assert.AreEqual(cY13data.Length, y13.Data.Length);
for (int i = 0; i < cY13data.Length; i++)
{
Assert.AreEqual(cY13data[i], y13.Data[i], delta);
}
//許容範囲を設定
delta = 0.0001;
//loss
Assert.AreEqual(cLoss.Data[0], loss.Data[0], delta);
//x.Grad
Assert.AreEqual(cXgrad.Length, x1.Grad.Length);
for (int i = 0; i < cXgrad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x1.Grad[i], delta);
}
Real[] cWgrad11 = Real.ToRealArray((Real[, ])cLinear1.W.Grad);
Real[] cbgrad11 = (Real[])cLinear1.b.Grad;
//W.grad
Assert.AreEqual(cWgrad11.Length, linear1.Weight.Grad.Length);
for (int i = 0; i < linear1.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad11[i], linear1.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad11.Length, linear1.Bias.Grad.Length);
for (int i = 0; i < linear1.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad11[i], linear1.Bias.Grad[i], delta);
}
Real[] cWgrad12 = Real.ToRealArray((Real[, ])cLinear2.W.Grad);
Real[] cbgrad12 = (Real[])cLinear2.b.Grad;
//W.grad
Assert.AreEqual(cWgrad12.Length, linear2.Weight.Grad.Length);
for (int i = 0; i < linear2.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad12[i], linear2.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad12.Length, linear2.Bias.Grad.Length);
for (int i = 0; i < linear2.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad12[i], linear2.Bias.Grad[i], delta);
}
//W.grad
int wLen = lstm.upward.Weight.Grad.Length;
Assert.AreEqual(cupwardWGrad.Length, lstm.upward.Weight.Grad.Length);
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
int bLen = lstm.upward.Bias.Length;
Assert.AreEqual(cupwardbGrad.Length, lstm.upward.Bias.Grad.Length);
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
//２周目
Variable <Real> cX2 = new Variable <Real>(Real.ToBaseNdArray(input2));
Variable <Real> cY21 = cLinear1.Forward(cX2);
Variable <Real> cY22 = cLstm.Forward(cY21);
Variable <Real> cY23 = cLinear2.Forward(cY22);
Variable <Real> cT2 = new Variable <Real>(Real.ToBaseNdArray(teach2));
Variable <Real> cLoss2 = new NChainer.MeanSquaredError <Real>().Forward(cY23, cT2);
//KelpNet
NdArray x2 = new NdArray(Real.ToRealArray(input2), new[] { 1 }, 5);
NdArray y21 = linear1.Forward(x2)[0];
NdArray y22 = lstm.Forward(y21)[0];
NdArray y23 = linear2.Forward(y22)[0];
NdArray t2 = new NdArray(Real.ToRealArray(teach2), new[] { 1 }, 5);
NdArray loss2 = new KelpNet.MeanSquaredError().Evaluate(y23, t2);
Assert.AreEqual(cLoss2.Data[0], loss2.Data[0], delta);
//Backwardを実行
cLoss2.Backward();
y23.Backward();
Real[] cYdata21 = Real.ToRealArray((Real[, ])cY21.Data);
Real[] cYdata22 = Real.ToRealArray((Real[, ])cY22.Data);
Real[] cYdata23 = Real.ToRealArray((Real[, ])cY23.Data);
Real[] cXgrad2 = Real.ToRealArray((Real[, ])cX2.Grad);
Real[] cupwardWGrad2 = Real.ToRealArray((Real[, ])cLstm.upward.W.Grad);
Real[] cupwardbGrad2 = (Real[])cLstm.upward.b.Grad;
Real[] clateralWGrad = Real.ToRealArray((Real[, ])cLstm.lateral.W.Grad);
//y21
Assert.AreEqual(cYdata21.Length, y21.Data.Length);
for (int i = 0; i < cYdata21.Length; i++)
{
Assert.AreEqual(cYdata21[i], y21.Data[i], delta);
}
//y22
Assert.AreEqual(cYdata22.Length, y22.Data.Length);
for (int i = 0; i < cYdata22.Length; i++)
{
Assert.AreEqual(cYdata22[i], y22.Data[i], delta);
}
//y23
Assert.AreEqual(cYdata23.Length, y23.Data.Length);
for (int i = 0; i < cYdata23.Length; i++)
{
Assert.AreEqual(cYdata23[i], y23.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad2.Length, x2.Grad.Length);
for (int i = 0; i < cXgrad2.Length; i++)
{
Assert.AreEqual(cXgrad2[i], x2.Grad[i], delta);
}
//経由が多くかなり誤差が大きい為
delta = 1.0;
Real[] cWgrad22 = Real.ToRealArray((Real[, ])cLinear2.W.Grad);
Real[] cbgrad22 = (Real[])cLinear2.b.Grad;
//W.grad
Assert.AreEqual(cWgrad22.Length, linear2.Weight.Grad.Length);
for (int i = 0; i < linear2.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad22[i], linear2.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad22.Length, linear2.Bias.Grad.Length);
for (int i = 0; i < linear2.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad22[i], linear2.Bias.Grad[i], delta);
}
delta = 2.0;
//W.grad
Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
for (int i = 0; i < clateralWGrad.Length; i++)
{
Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
}
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad2[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad2[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
delta = 20.0;
Real[] cWgrad21 = Real.ToRealArray((Real[, ])cLinear1.W.Grad);
Real[] cbgrad21 = (Real[])cLinear1.b.Grad;
//W.grad
Assert.AreEqual(cWgrad21.Length, linear1.Weight.Grad.Length);
for (int i = 0; i < linear1.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad21[i], linear1.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad21.Length, linear1.Bias.Grad.Length);
for (int i = 0; i < linear1.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad21[i], linear1.Bias.Grad[i], delta);
}
}
public void LRNRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int n = Mother.Dice.Next(2, 7);
float k = (float)Mother.Dice.NextDouble() * 3;
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int width = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
Real[,,,] input = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, width);
Real[,,,] dummyGy = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, width);
//chainer
LocalResponseNormalization <Real> cLocalResponseNormalization = new LocalResponseNormalization <Real>(n, k);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cLocalResponseNormalization.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//kelpnet
LRN <Real> lrn = new LRN <Real>(n, k);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = lrn.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, , , ])cY.Data).Flatten();
Real[] cXgrad = ((Real[, , , ])cX.Grad).Flatten();
//許容範囲を算出
Real delta = 0.00001f;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void SwishRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int ioCount = Mother.Dice.Next(1, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, ioCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, ioCount);
Real beta = (Real)Mother.Dice.NextDouble();
//Chainer
NChainer.Swish <Real> cSwish = new NChainer.Swish <Real>(new[] { ioCount }, beta);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cSwish.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
Swish <Real> swish = new Swish <Real>(new[] { ioCount }, beta);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = swish.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, ])cY.Data).Flatten();
Real[] cXgrad = ((Real[, ])cX.Grad).Flatten();
Real[] cbgrad = (Real[])cSwish.beta.Grad;
//許容範囲を算出
Real delta = 0.00001f;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad.Length, swish.Beta.Grad.Length);
for (int i = 0; i < swish.Beta.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], swish.Beta.Grad[i], delta);
}
}
public void TrainTest(bool isTtrain, bool finetune)
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 50);
int ioCount = Mother.Dice.Next(1, 50);
Real[,] input = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, ioCount });
Real[,] dummyGy = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, ioCount });
Real[] gamma = Initializer.GetRealArray(ioCount);
Real[] beta = Initializer.GetRealArray(ioCount);
Real[] avgMean = Initializer.GetRealArray(ioCount);
Real[] avgVar = Initializer.GetRealArray(ioCount);
//Chainer
Chainer.Config["train"] = isTtrain;
NChainer.BatchNormalization <Real> cBatchNormalization = new NChainer.BatchNormalization <Real>(ioCount, dtype: Real.Type);
cBatchNormalization.gamma = new Variable <Real>(Real.ToBaseNdArray(gamma));
cBatchNormalization.beta = new Variable <Real>(Real.ToBaseNdArray(beta));
cBatchNormalization.avgMean = Real.ToBaseNdArray(avgMean);
cBatchNormalization.avgVar = Real.ToBaseNdArray(avgVar);
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cBatchNormalization.Forward(cX, finetune);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.BatchNormalization batchNormalization = new BatchNormalization(ioCount, train: isTtrain, finetune: finetune);
batchNormalization.Gamma.Data = Real.ToRealArray(gamma);
batchNormalization.Beta.Data = Real.ToRealArray(beta);
batchNormalization.AvgMean.Data = Real.ToRealArray(avgMean);
batchNormalization.AvgVar.Data = Real.ToRealArray(avgVar);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { ioCount }, batchCount);
NdArray y = batchNormalization.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, ])cY.Data);
Real[] cXgrad = Real.ToRealArray((Real[, ])cX.Grad);
Real[] cGammaGrad = Real.ToRealArray((Real[])cBatchNormalization.gamma.Grad);
Real[] cBetaGrad = Real.ToRealArray((Real[])cBatchNormalization.beta.Grad);
//許容範囲を算出
double delta = 0.00001;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//gamma.grad
Assert.AreEqual(cGammaGrad.Length, batchNormalization.Gamma.Grad.Length);
for (int i = 0; i < batchNormalization.Gamma.Grad.Length; i++)
{
Assert.AreEqual(cGammaGrad[i], batchNormalization.Gamma.Grad[i], delta);
}
//beta.grad
Assert.AreEqual(cBetaGrad.Length, batchNormalization.Beta.Grad.Length);
for (int i = 0; i < batchNormalization.Beta.Grad.Length; i++)
{
Assert.AreEqual(cBetaGrad[i], batchNormalization.Beta.Grad[i], delta);
}
}
public void ConcatRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int widthA = Mother.Dice.Next(1, 16);
int widthB = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
int axis = 3;
Real[,,,] inputA = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, widthA });
Real[,,,] inputB = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, widthB });
Real[,,,] dummyGy = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, widthA + widthB });
//chainer
NChainer.Concat <Real> cConcat = new NChainer.Concat <Real>(axis);
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(inputA));
Variable <Real> cY = new Variable <Real>(Real.ToBaseNdArray(inputB));
Variable <Real> cZ = cConcat.Forward(cX, cY);
cZ.Grad = Real.ToBaseNdArray(dummyGy);
cZ.Backward();
//KelpNet
KelpNet.Concat concat = new Concat(axis - 1);//Chainerと異なり1次元目を暗黙的にBatchとみなさないため
NdArray x = new NdArray(Real.ToRealArray(inputA), new[] { ch, height, widthA }, batchCount);
NdArray y = new NdArray(Real.ToRealArray(inputB), new[] { ch, height, widthB }, batchCount);
NdArray z = concat.Forward(x, y)[0];
z.Grad = Real.ToRealArray(dummyGy);
z.Backward();
Real[] cZdata = Real.ToRealArray((Real[, , , ])cZ.Data);
//Copyが必要
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad.Copy());
Real[] cYgrad = Real.ToRealArray((Real[, , , ])cY.Grad.Copy());
//許容範囲を算出
double delta = 0.00001;
//z
Assert.AreEqual(cZdata.Length, z.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cZdata[i], z.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//y.grad
Assert.AreEqual(cYgrad.Length, y.Grad.Length);
for (int i = 0; i < y.Grad.Length; i++)
{
Assert.AreEqual(cYgrad[i], y.Grad[i], delta);
}
}
public void AVGPoolingRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int chCount = Mother.Dice.Next(1, 5);
int wideSize = Mother.Dice.Next(8, 32);
int heightSize = Mother.Dice.Next(8, 32);
int kWidth = Mother.Dice.Next(1, 5);
int kHeight = Mother.Dice.Next(1, 5);
int strideX = Mother.Dice.Next(1, 5);
int strideY = Mother.Dice.Next(1, 5);
int padX = Mother.Dice.Next(0, 5);
int padY = Mother.Dice.Next(0, 5);
int outputHeight = (int)Math.Floor((heightSize - kHeight + padY * 2.0f) / strideY) + 1;
int outputWidth = (int)Math.Floor((wideSize - kWidth + padX * 2.0f) / strideX) + 1;
Real[,,,] input = Initializer.GetRandomValues <Real[, , , ]>(batchCount, chCount, heightSize, wideSize);
Real[,,,] dummyGy = Initializer.GetRandomValues <Real[, , , ]>(batchCount, chCount, outputHeight, outputWidth);
//Chainer
NChainer.AveragePooling2D <Real> cMaxPooling2D = new NChainer.AveragePooling2D <Real>(
new[] { kHeight, kWidth },
new[] { strideY, strideX },
new[] { padY, padX }
);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cMaxPooling2D.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
KelpNet.AveragePooling2D <Real> maxPooling2D = new KelpNet.AveragePooling2D <Real>(
new[] { kWidth, kHeight },
new[] { strideX, strideY },
new[] { padX, padY });
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = maxPooling2D.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, , , ])cY.Data.Copy()).Flatten();
Real[] cXgrad = ((Real[, , , ])cX.Grad.Copy()).Flatten();
//許容範囲を算出
Real delta = 0.00001f;
Assert.AreEqual(cYdata.Length, y.Data.Length);
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
//y
for (int i = 0; i < y.Data.Length; i++)
{
if (cYdata[i] < float.Epsilon && cYdata[i] > -float.Epsilon)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
}
//x.grad
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void AdamRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 50);
int outputCount = Mother.Dice.Next(2, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, inputCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, outputCount);
Real[,] w = Initializer.GetRandomValues <Real[, ]>(outputCount, inputCount);
Real[] b = Initializer.GetRandomValues <Real[]>(outputCount);
float alpha = (float)Mother.Dice.NextDouble(); //0.001f
float beta1 = (float)Mother.Dice.NextDouble(); //0.9f;
float beta2 = (float)Mother.Dice.NextDouble(); //0.999f;
float eps = (float)Mother.Dice.NextDouble(); //1e-08f;
float eta = (float)Mother.Dice.NextDouble(); //1.0f;
//Chainer
NChainer.Linear <Real> cLinear = new NChainer.Linear <Real>(inputCount, outputCount, false, w, b);
NChainer.Adam <Real> cAdam = new NChainer.Adam <Real>(alpha, beta1, beta2, eps, eta);
cAdam.Setup(cLinear);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cLinear.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
cAdam.Update();
//KelpNet
KelpNet.CL.Linear <Real> linear = new KelpNet.CL.Linear <Real>(inputCount, outputCount, false, w, b);
KelpNet.Adam <Real> adam = new Adam <Real>(alpha, beta1, beta2, eps, eta);
adam.SetUp(linear);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = linear.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
adam.Update();
Real[] cW = ((Real[, ])cLinear.W.Data).Flatten();
Real[] cb = (Real[])cLinear.b.Data;
//許容範囲を算出
Real delta = 0.00001f;
//W.grad
Assert.AreEqual(cW.Length, linear.Weight.Data.Length);
for (int i = 0; i < linear.Weight.Data.Length; i++)
{
Assert.AreEqual(cW[i], linear.Weight.Data[i], delta);
}
//b.grad
Assert.AreEqual(cb.Length, linear.Bias.Data.Length);
for (int i = 0; i < linear.Bias.Data.Length; i++)
{
Assert.AreEqual(cb[i], linear.Bias.Data[i], delta);
}
}
public void SGDRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 50);
int outputCount = Mother.Dice.Next(2, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, inputCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, outputCount);
Real[,] w = Initializer.GetRandomValues <Real[, ]>(outputCount, inputCount);
Real[] b = Initializer.GetRandomValues <Real[]>(outputCount);
//Chainer
Linear <Real> cLinear = new Linear <Real>(inputCount, outputCount, false, w, b);
NChainer.SGD <Real> cSgd = new NChainer.SGD <Real>();
cSgd.Setup(cLinear);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cLinear.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
cSgd.Update();
//KelpNet
CL.Linear <Real> linear = new CL.Linear <Real>(inputCount, outputCount, false, w, b);
KelpNet.SGD <Real> sgd = new SGD <Real>();
sgd.SetUp(linear);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = linear.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
sgd.Update();
Real[] cW = ((Real[, ])cLinear.W.Data).Flatten();
Real[] cb = (Real[])cLinear.b.Data;
//許容範囲を算出
Real delta = 0.00001f;
//W.grad
Assert.AreEqual(cW.Length, linear.Weight.Data.Length);
for (int i = 0; i < linear.Weight.Data.Length; i++)
{
Assert.AreEqual(cW[i], linear.Weight.Data[i], delta);
}
//b.grad
Assert.AreEqual(cb.Length, linear.Bias.Data.Length);
for (int i = 0; i < linear.Bias.Data.Length; i++)
{
Assert.AreEqual(cb[i], linear.Bias.Data[i], delta);
}
}
public void MeanSquaredRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int width = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
Real[,,,] inputA = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, width });
Real[,,,] inputB = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, width });
for (int i = 0; i < inputB.GetLength(0); i++)
{
for (int j = 0; j < inputB.GetLength(1); j++)
{
for (int k = 0; k < inputB.GetLength(2); k++)
{
for (int l = 0; l < inputB.GetLength(3); l++)
{
inputB[i, j, k, l] *= 3.1415;
}
}
}
}
//chainer
NChainer.MeanSquaredError <Real> cMeanSquaredError = new NChainer.MeanSquaredError <Real>();
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(inputA));
Variable <Real> cY = new Variable <Real>(Real.ToBaseNdArray(inputB));
Variable <Real> cZ = cMeanSquaredError.Forward(cX, cY);
cZ.Backward();
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad);
//KelpNet
KelpNet.MeanSquaredError meanSquaredError = new KelpNet.MeanSquaredError();
NdArray x = new NdArray(Real.ToRealArray(inputA), new[] { ch, height, width }, batchCount);
NdArray y = new NdArray(Real.ToRealArray(inputB), new[] { ch, height, width }, batchCount);
//KelpNetはBackwaward側のみEvaluateで実行される
NdArray z = meanSquaredError.Evaluate(x, y);
//許容範囲を算出(内部の割引順が違うため誤差が大きい)
double delta = 0.001;
//Loss
Assert.AreEqual(cZ.Data[0], z.Data[0], delta);
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void LRNRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int n = Mother.Dice.Next(2, 7);
float k = (float)Mother.Dice.NextDouble() * 3;
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int width = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
Real[,,,] input = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, width });
Real[,,,] dummyGy = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, ch, height, width });
//chainer
NChainer.LocalResponseNormalization <Real> cLocalResponseNormalization = new NChainer.LocalResponseNormalization <Real>(n, k);
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cLocalResponseNormalization.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//kelpnet
KelpNet.LRN lrn = new LRN(n, k);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { ch, height, width }, batchCount);
NdArray y = lrn.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, , , ])cY.Data);
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad);
//許容範囲を算出
double delta = 0.00001;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void SplitAxisRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int widthA = Mother.Dice.Next(1, 16);
int widthB = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
int axis = 3;
Real[,,,] input = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthA + widthB);
Real[,,,] dummyGyA = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthA);
Real[,,,] dummyGyB = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, widthB);
//chainer
NChainer.SplitAxis <Real> cSplitAxis = new NChainer.SplitAxis <Real>();
Variable <Real> cX = new Variable <Real>(input);
PyObject[] cY = cSplitAxis.Forward(cX, new[] { widthA }, axis);
Variable <Real> cY0 = cY[0];
Variable <Real> cY1 = cY[1];
cY0.Grad = dummyGyA;
cY1.Grad = dummyGyB;
//Chainerはどちらか一方で両方分のBackwardが走る
cY0.Backward();
//cY1.Backward();
//KelpNet
SplitAxis <Real> splitAxis = new SplitAxis <Real>(new[] { widthA }, axis - 1); //Chainerと異なり1次元目を暗黙的にBatchとみなさないため
NdArray <Real> x = new NdArray <Real>(input, asBatch: true); // new NdArray(Real.ToRealArray(), new[] { ch, height, widthA + widthB }, batchCount);
NdArray <Real>[] y = splitAxis.Forward(x);
y[0].Grad = dummyGyA.Flatten();
y[1].Grad = dummyGyB.Flatten();
//KelpNetは出力した両方からBackwardしないと処理が走らない
y[0].Backward();
y[1].Backward();
//Copyが必要
Real[] cY0data = ((Real[, , , ])cY0.Data.Copy()).Flatten();
Real[] cY1data = ((Real[, , , ])cY1.Data.Copy()).Flatten();
Real[] cXgrad = ((Real[, , , ])cX.Grad).Flatten();
//許容範囲を算出
Real delta = 0.00001f;
//y0
Assert.AreEqual(cY0data.Length, y[0].Data.Length);
for (int i = 0; i < y[0].Data.Length; i++)
{
Assert.AreEqual(cY0data[i], y[0].Data[i], delta);
}
//y1
Assert.AreEqual(cY1data.Length, y[1].Data.Length);
for (int i = 0; i < y[1].Data.Length; i++)
{
Assert.AreEqual(cY1data[i], y[1].Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void RandomTest(bool gpuEnable)
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(1, 50);
int outputCount = Mother.Dice.Next(1, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, inputCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, outputCount);
Real[,] w = Initializer.GetRandomValues <Real[, ]>(outputCount, inputCount);
Real[] b = Initializer.GetRandomValues <Real[]>(outputCount);
//Chainer
NChainer.Linear <Real> cLinear = new NChainer.Linear <Real>(inputCount, outputCount, false, w, b);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cLinear.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
CL.Linear <Real> linear = new CL.Linear <Real>(inputCount, outputCount, false, w, b, gpuEnable: gpuEnable);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = linear.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, ])cY.Data).Flatten();
Real[] cXgrad = ((Real[, ])cX.Grad).Flatten();
Real[] cWgrad = ((Real[, ])cLinear.W.Grad).Flatten();
Real[] cbgrad = (Real[])cLinear.b.Grad;
//許容範囲を算出
Real delta = 0.00001f;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//W.grad
Assert.AreEqual(cWgrad.Length, linear.Weight.Grad.Length);
for (int i = 0; i < linear.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], linear.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad.Length, linear.Bias.Grad.Length);
for (int i = 0; i < linear.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], linear.Bias.Grad[i], delta);
}
}
public void RandomTest(bool gpuEnable)
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int inChCount = Mother.Dice.Next(1, 5);
int outChCount = Mother.Dice.Next(1, 5);
int wideSize = Mother.Dice.Next(8, 32);
int heightSize = Mother.Dice.Next(8, 32);
int kWidth = Mother.Dice.Next(1, 5);
int kHeight = Mother.Dice.Next(1, 5);
int strideX = Mother.Dice.Next(1, 5);
int strideY = Mother.Dice.Next(1, 5);
int padX = Mother.Dice.Next(0, 5);
int padY = Mother.Dice.Next(0, 5);
int outputHeight = (int)Math.Floor((heightSize - kHeight + padY * 2.0f) / strideY) + 1;
int outputWidth = (int)Math.Floor((wideSize - kWidth + padX * 2.0f) / strideX) + 1;
Real[,,,] input = Initializer.GetRandomValues <Real[, , , ]>(batchCount, inChCount, heightSize, wideSize);
Real[,,,] dummyGy = Initializer.GetRandomValues <Real[, , , ]>(batchCount, outChCount, outputHeight, outputWidth);
Real[,,,] w = Initializer.GetRandomValues <Real[, , , ]>(outChCount, inChCount, kHeight, kWidth);
Real[] b = Initializer.GetRandomValues <Real[]>(outChCount);
//Chainer
NChainer.Convolution2D <Real> cConvolotion2D = new NChainer.Convolution2D <Real>(
inChCount, outChCount,
new[] { kHeight, kWidth },
new[] { strideY, strideX },
new[] { padY, padX },
false,
w,
b);
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cConvolotion2D.Forward(cX);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
CL.Convolution2D <Real> convolution2D = new CL.Convolution2D <Real>(
inChCount, outChCount,
new[] { kWidth, kHeight },
new[] { strideX, strideY },
new[] { padX, padY },
false, w, b, gpuEnable: gpuEnable);
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = convolution2D.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, , , ])cY.Data.Copy()).Flatten();
Real[] cXgrad = ((Real[, , , ])cX.Grad.Copy()).Flatten();
Real[] cWgrad = ((Real[, , , ])cConvolotion2D.W.Grad).Flatten();
Real[] cbgrad = (Real[])cConvolotion2D.b.Grad;
//許容範囲を算出
Real delta = 0.00001f;
Assert.AreEqual(cYdata.Length, y.Data.Length);
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
Assert.AreEqual(cWgrad.Length, convolution2D.Weight.Grad.Length);
Assert.AreEqual(cbgrad.Length, convolution2D.Bias.Grad.Length);
//y
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
delta = 0.1f;
//W.grad
for (int i = 0; i < convolution2D.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad[i], convolution2D.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < convolution2D.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad[i], convolution2D.Bias.Grad[i], delta);
}
}
public void RnnLSTMRandomTest()
{
Python.Initialize();
Chainer.Initialize();
Real[,] input = { { 1.0f }, { 3.0f }, { 5.0f }, { 7.0f }, { 9.0f } };
Real[,] teach = { { 3.0f }, { 5.0f }, { 7.0f }, { 9.0f }, { 11.0f } };
Real[,] input2 = { { 3.0f }, { 5.0f }, { 7.0f }, { 9.0f }, { 11.0f } };
Real[,] teach2 = { { 5.0f }, { 7.0f }, { 9.0f }, { 11.0f }, { 13.0f } };
int outputCount = 1;
int inputCount = 1;
int hiddenCount = 2;
Real[,] upwardInit = Initializer.GetRandomValues <Real[, ]>(hiddenCount, hiddenCount);
Real[,] lateralInit = Initializer.GetRandomValues <Real[, ]>(hiddenCount, hiddenCount);
Real[,,] biasInit = Initializer.GetRandomValues <Real[, , ]>(1, hiddenCount, 1);
Real[,,] forgetBiasInit = Initializer.GetRandomValues <Real[, , ]>(1, hiddenCount, 1);
//Chainer
Real[,] w1 = Initializer.GetRandomValues <Real[, ]>(hiddenCount, inputCount);
Real[] b1 = Initializer.GetRandomValues <Real[]>(hiddenCount);
//Chainer
Linear <Real> cLinear1 = new Linear <Real>(inputCount, hiddenCount, false, w1, b1);
NChainer.LSTM <Real> cLstm = new NChainer.LSTM <Real>(hiddenCount, hiddenCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
Real[,] w2 = Initializer.GetRandomValues <Real[, ]>(outputCount, hiddenCount);
Real[] b2 = Initializer.GetRandomValues <Real[]>(outputCount);
Linear <Real> cLinear2 = new Linear <Real>(hiddenCount, outputCount, false, w2, b2);
Variable <Real> cX1 = new Variable <Real>(input);
Variable <Real> cY11 = cLinear1.Forward(cX1);
Variable <Real> cY12 = cLstm.Forward(cY11);
Variable <Real> cY13 = cLinear2.Forward(cY12);
Variable <Real> cT = new Variable <Real>(teach);
Variable <Real> cLoss = new NChainer.MeanSquaredError <Real>().Forward(cY13, cT);
cLoss.Backward();
//KelpNet
CL.Linear <Real> linear1 = new CL.Linear <Real>(inputCount, hiddenCount, false, w1, b1);
LSTM <Real> lstm = new LSTM <Real>(hiddenCount, hiddenCount, lateralInit, upwardInit, biasInit, forgetBiasInit);
CL.Linear <Real> linear2 = new CL.Linear <Real>(hiddenCount, outputCount, false, w2, b2);
NdArray <Real> x1 = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y11 = linear1.Forward(x1)[0];
NdArray <Real> y12 = lstm.Forward(y11)[0];
NdArray <Real> y13 = linear2.Forward(y12)[0];
NdArray <Real> t = new NdArray <Real>(teach, asBatch: true);
NdArray <Real> loss = new MeanSquaredError <Real>().Evaluate(y13, t);
y13.Backward();
Real[] cY11data = ((Real[, ])cY11.Data).Flatten();
Real[] cY12data = ((Real[, ])cY12.Data).Flatten();
Real[] cY13data = ((Real[, ])cY13.Data).Flatten();
Real[] cXgrad = ((Real[, ])cX1.Grad).Flatten();
Real[] cupwardWGrad = ((Real[, ])cLstm.upward.W.Grad).Flatten();
Real[] cupwardbGrad = (Real[])cLstm.upward.b.Grad;
//許容範囲を設定
Real delta = 0.00001f;
//y11
Assert.AreEqual(cY11data.Length, y11.Data.Length);
for (int i = 0; i < cY11data.Length; i++)
{
Assert.AreEqual(cY11data[i], y11.Data[i], delta);
}
//y12
Assert.AreEqual(cY12data.Length, y12.Data.Length);
for (int i = 0; i < cY12data.Length; i++)
{
Assert.AreEqual(cY12data[i], y12.Data[i], delta);
}
//y13
Assert.AreEqual(cY13data.Length, y13.Data.Length);
for (int i = 0; i < cY13data.Length; i++)
{
Assert.AreEqual(cY13data[i], y13.Data[i], delta);
}
//許容範囲を設定
delta = 0.0001f;
//loss
Assert.AreEqual(cLoss.Data[0], loss.Data[0], delta);
//x.Grad
Assert.AreEqual(cXgrad.Length, x1.Grad.Length);
for (int i = 0; i < cXgrad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x1.Grad[i], delta);
}
Real[] cWgrad11 = ((Real[, ])cLinear1.W.Grad).Flatten();
Real[] cbgrad11 = (Real[])cLinear1.b.Grad;
//W.grad
Assert.AreEqual(cWgrad11.Length, linear1.Weight.Grad.Length);
for (int i = 0; i < linear1.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad11[i], linear1.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad11.Length, linear1.Bias.Grad.Length);
for (int i = 0; i < linear1.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad11[i], linear1.Bias.Grad[i], delta);
}
Real[] cWgrad12 = ((Real[, ])cLinear2.W.Grad).Flatten();
Real[] cbgrad12 = (Real[])cLinear2.b.Grad;
//W.grad
Assert.AreEqual(cWgrad12.Length, linear2.Weight.Grad.Length);
for (int i = 0; i < linear2.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad12[i], linear2.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad12.Length, linear2.Bias.Grad.Length);
for (int i = 0; i < linear2.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad12[i], linear2.Bias.Grad[i], delta);
}
//W.grad
int wLen = lstm.upward.Weight.Grad.Length;
Assert.AreEqual(cupwardWGrad.Length, lstm.upward.Weight.Grad.Length);
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
int bLen = lstm.upward.Bias.Length;
Assert.AreEqual(cupwardbGrad.Length, lstm.upward.Bias.Grad.Length);
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
//２周目
Variable <Real> cX2 = new Variable <Real>(input2);
Variable <Real> cY21 = cLinear1.Forward(cX2);
Variable <Real> cY22 = cLstm.Forward(cY21);
Variable <Real> cY23 = cLinear2.Forward(cY22);
Variable <Real> cT2 = new Variable <Real>(teach2);
Variable <Real> cLoss2 = new NChainer.MeanSquaredError <Real>().Forward(cY23, cT2);
//KelpNet
NdArray <Real> x2 = new NdArray <Real>(input2, asBatch: true);
NdArray <Real> y21 = linear1.Forward(x2)[0];
NdArray <Real> y22 = lstm.Forward(y21)[0];
NdArray <Real> y23 = linear2.Forward(y22)[0];
NdArray <Real> t2 = new NdArray <Real>(teach2, asBatch: true);
NdArray <Real> loss2 = new MeanSquaredError <Real>().Evaluate(y23, t2);
Assert.AreEqual(cLoss2.Data[0], loss2.Data[0], delta);
//Backwardを実行
cLoss2.Backward();
y23.Backward();
Real[] cYdata21 = ((Real[, ])cY21.Data).Flatten();
Real[] cYdata22 = ((Real[, ])cY22.Data).Flatten();
Real[] cYdata23 = ((Real[, ])cY23.Data).Flatten();
Real[] cXgrad2 = ((Real[, ])cX2.Grad).Flatten();
Real[] cupwardWGrad2 = ((Real[, ])cLstm.upward.W.Grad).Flatten();
Real[] cupwardbGrad2 = (Real[])cLstm.upward.b.Grad;
Real[] clateralWGrad = ((Real[, ])cLstm.lateral.W.Grad).Flatten();
//y21
Assert.AreEqual(cYdata21.Length, y21.Data.Length);
for (int i = 0; i < cYdata21.Length; i++)
{
Assert.AreEqual(cYdata21[i], y21.Data[i], delta);
}
//y22
Assert.AreEqual(cYdata22.Length, y22.Data.Length);
for (int i = 0; i < cYdata22.Length; i++)
{
Assert.AreEqual(cYdata22[i], y22.Data[i], delta);
}
//y23
Assert.AreEqual(cYdata23.Length, y23.Data.Length);
for (int i = 0; i < cYdata23.Length; i++)
{
Assert.AreEqual(cYdata23[i], y23.Data[i], delta);
}
//x.Grad
Assert.AreEqual(cXgrad2.Length, x2.Grad.Length);
for (int i = 0; i < cXgrad2.Length; i++)
{
Assert.AreEqual(cXgrad2[i], x2.Grad[i], delta);
}
//経由が多くかなり誤差が大きい為
delta = 1.0f;
Real[] cWgrad22 = ((Real[, ])cLinear2.W.Grad).Flatten();
Real[] cbgrad22 = (Real[])cLinear2.b.Grad;
//W.grad
Assert.AreEqual(cWgrad22.Length, linear2.Weight.Grad.Length);
for (int i = 0; i < linear2.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad22[i], linear2.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad22.Length, linear2.Bias.Grad.Length);
for (int i = 0; i < linear2.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad22[i], linear2.Bias.Grad[i], delta);
}
delta = 2.0f;
//W.grad
Assert.AreEqual(clateralWGrad.Length, lstm.lateral.Weight.Grad.Length);
for (int i = 0; i < clateralWGrad.Length; i++)
{
Assert.AreEqual(clateralWGrad[i + wLen * 0], lstm.lateral.Weight.Grad[i], delta);
}
for (int i = 0; i < wLen; i++)
{
Assert.AreEqual(cupwardWGrad2[i + wLen * 0], lstm.upward.Weight.Grad[i], delta);
}
//b.grad
for (int i = 0; i < bLen; i++)
{
Assert.AreEqual(cupwardbGrad2[i + wLen * 0], lstm.upward.Bias.Grad[i], delta);
}
delta = 20.0f;
Real[] cWgrad21 = ((Real[, ])cLinear1.W.Grad).Flatten();
Real[] cbgrad21 = (Real[])cLinear1.b.Grad;
//W.grad
Assert.AreEqual(cWgrad21.Length, linear1.Weight.Grad.Length);
for (int i = 0; i < linear1.Weight.Grad.Length; i++)
{
Assert.AreEqual(cWgrad21[i], linear1.Weight.Grad[i], delta);
}
//b.grad
Assert.AreEqual(cbgrad21.Length, linear1.Bias.Grad.Length);
for (int i = 0; i < linear1.Bias.Grad.Length; i++)
{
Assert.AreEqual(cbgrad21[i], linear1.Bias.Grad[i], delta);
}
}
public void RandomTest(bool gpuEnable)
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int chCount = Mother.Dice.Next(1, 5);
int wideSize = Mother.Dice.Next(8, 32);
int heightSize = Mother.Dice.Next(8, 32);
int kWidth = Mother.Dice.Next(1, 5);
int kHeight = Mother.Dice.Next(1, 5);
int strideX = Mother.Dice.Next(1, 5);
int strideY = Mother.Dice.Next(1, 5);
int padX = Mother.Dice.Next(0, 5);
int padY = Mother.Dice.Next(0, 5);
bool coverAll = Mother.Dice.Next(1) == 0;
int outputHeight = coverAll ?
(int)Math.Floor((heightSize - kHeight + padY * 2.0 + strideY - 1) / strideY) + 1 :
(int)Math.Floor((heightSize - kHeight + padY * 2.0) / strideY) + 1;
int outputWidth = coverAll ?
(int)Math.Floor((wideSize - kWidth + padX * 2.0 + strideX - 1) / strideX) + 1 :
(int)Math.Floor((wideSize - kWidth + padX * 2.0) / strideX) + 1;
Real[,,,] input = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, chCount, heightSize, wideSize });
Real[,,,] dummyGy = (Real[, , , ])Initializer.GetRealNdArray(new[] { batchCount, chCount, outputHeight, outputWidth });
//Chainer
NChainer.MaxPooling2D <Real> cMaxPooling2D = new NChainer.MaxPooling2D <Real>(
new[] { kHeight, kWidth },
new[] { strideY, strideX },
new[] { padY, padX }
);
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cMaxPooling2D.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
//KelpNet
KelpNet.CL.MaxPooling2D maxPooling2D = new KelpNet.CL.MaxPooling2D(
new[] { kWidth, kHeight },
new[] { strideX, strideY },
new[] { padX, padY },
gpuEnable: gpuEnable);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { chCount, heightSize, wideSize }, batchCount);
NdArray y = maxPooling2D.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
Real[] cYdata = Real.ToRealArray((Real[, , , ])cY.Data.Copy());
Real[] cXgrad = Real.ToRealArray((Real[, , , ])cX.Grad.Copy());
//許容範囲を算出
double delta = 0.00001;
Assert.AreEqual(cYdata.Length, y.Data.Length);
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
//y
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void SGDRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int inputCount = Mother.Dice.Next(2, 50);
int outputCount = Mother.Dice.Next(2, 50);
int batchCount = Mother.Dice.Next(1, 5);
Real[,] input = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, inputCount });
Real[,] dummyGy = (Real[, ])Initializer.GetRealNdArray(new[] { batchCount, outputCount });
Real[,] w = (Real[, ])Initializer.GetRealNdArray(new[] { outputCount, inputCount });
Real[] b = Initializer.GetRealArray(outputCount);
//Chainer
NChainer.Linear <Real> cLinear = new NChainer.Linear <Real>(inputCount, outputCount, false, Real.ToBaseNdArray(w), Real.ToBaseArray(b));
NChainer.SGD <Real> cSgd = new NChainer.SGD <Real>();
cSgd.Setup(cLinear);
Variable <Real> cX = new Variable <Real>(Real.ToBaseNdArray(input));
Variable <Real> cY = cLinear.Forward(cX);
cY.Grad = Real.ToBaseNdArray(dummyGy);
cY.Backward();
cSgd.Update();
//KelpNet
KelpNet.Linear linear = new KelpNet.Linear(inputCount, outputCount, false, w, b);
KelpNet.SGD sgd = new SGD();
sgd.SetUp(linear);
NdArray x = new NdArray(Real.ToRealArray(input), new[] { inputCount }, batchCount);
NdArray y = linear.Forward(x)[0];
y.Grad = Real.ToRealArray(dummyGy);
y.Backward();
sgd.Update();
Real[] cW = Real.ToRealArray((Real[, ])cLinear.W.Data);
Real[] cb = (Real[])cLinear.b.Data;
//許容範囲を算出
double delta = 0.00001;
//W.grad
Assert.AreEqual(cW.Length, linear.Weight.Data.Length);
for (int i = 0; i < linear.Weight.Data.Length; i++)
{
Assert.AreEqual(cW[i], linear.Weight.Data[i], delta);
}
//b.grad
Assert.AreEqual(cb.Length, linear.Bias.Data.Length);
for (int i = 0; i < linear.Bias.Data.Length; i++)
{
Assert.AreEqual(cb[i], linear.Bias.Data[i], delta);
}
}
public void MeanSquaredRandomTest()
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 5);
int ch = Mother.Dice.Next(1, 5);
int width = Mother.Dice.Next(1, 16);
int height = Mother.Dice.Next(1, 16);
Real[,,,] inputA = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, width);
Real[,,,] inputB = Initializer.GetRandomValues <Real[, , , ]>(batchCount, ch, height, width);
for (int i = 0; i < inputB.GetLength(0); i++)
{
for (int j = 0; j < inputB.GetLength(1); j++)
{
for (int k = 0; k < inputB.GetLength(2); k++)
{
for (int l = 0; l < inputB.GetLength(3); l++)
{
inputB[i, j, k, l] *= (Real)3.1415f;
}
}
}
}
//chainer
NChainer.MeanSquaredError <Real> cMeanSquaredError = new NChainer.MeanSquaredError <Real>();
Variable <Real> cX = new Variable <Real>(inputA);
Variable <Real> cY = new Variable <Real>(inputB);
Variable <Real> cZ = cMeanSquaredError.Forward(cX, cY);
cZ.Backward();
Real[] cXgrad = ((Real[, , , ])cX.Grad).Flatten();
//KelpNet
KelpNet.MeanSquaredError <Real> meanSquaredError = new KelpNet.MeanSquaredError <Real>();
NdArray <Real> x = new NdArray <Real>(inputA, asBatch: true);
NdArray <Real> y = new NdArray <Real>(inputB, asBatch: true);
//KelpNetはBackwaward側のみEvaluateで実行される
NdArray <Real> z = meanSquaredError.Evaluate(x, y);
//許容範囲を算出(内部の割引順が違うため誤差が大きい)
Real delta = 0.001f;
//Loss
Assert.AreEqual(cZ.Data[0], z.Data[0], delta);
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
}
public void TrainTest(bool isTtrain, bool finetune)
{
Python.Initialize();
Chainer.Initialize();
int batchCount = Mother.Dice.Next(1, 50);
int ioCount = Mother.Dice.Next(1, 50);
Real[,] input = Initializer.GetRandomValues <Real[, ]>(batchCount, ioCount);
Real[,] dummyGy = Initializer.GetRandomValues <Real[, ]>(batchCount, ioCount);
Real[] gamma = Initializer.GetRandomValues <Real[]>(ioCount);
Real[] beta = Initializer.GetRandomValues <Real[]>(ioCount);
Real[] avgMean = Initializer.GetRandomValues <Real[]>(ioCount);
Real[] avgVar = Initializer.GetRandomValues <Real[]>(ioCount);
//Chainer
Chainer.Config["train"] = isTtrain;
NChainer.BatchNormalization <Real> cBatchNormalization = new NChainer.BatchNormalization <Real>(ioCount, dtype: typeof(Real));
cBatchNormalization.gamma = new Variable <Real>(gamma);
cBatchNormalization.beta = new Variable <Real>(beta);
cBatchNormalization.avgMean = avgMean;
cBatchNormalization.avgVar = avgVar;
Variable <Real> cX = new Variable <Real>(input);
Variable <Real> cY = cBatchNormalization.Forward(cX, finetune);
cY.Grad = dummyGy;
cY.Backward();
//KelpNet
KelpNet.BatchNormalization <Real> batchNormalization = new BatchNormalization <Real>(ioCount, train: isTtrain, finetune: finetune);
batchNormalization.Gamma.Data = gamma;
batchNormalization.Beta.Data = beta;
batchNormalization.AvgMean.Data = avgMean;
batchNormalization.AvgVar.Data = avgVar;
NdArray <Real> x = new NdArray <Real>(input, asBatch: true);
NdArray <Real> y = batchNormalization.Forward(x)[0];
y.Grad = dummyGy.Flatten();
y.Backward();
Real[] cYdata = ((Real[, ])cY.Data).Flatten();
Real[] cXgrad = ((Real[, ])cX.Grad).Flatten();
Real[] cGammaGrad = (Real[])cBatchNormalization.gamma.Grad;
Real[] cBetaGrad = (Real[])cBatchNormalization.beta.Grad;
//許容範囲を算出
Real delta = 0.00005f;
//y
Assert.AreEqual(cYdata.Length, y.Data.Length);
for (int i = 0; i < y.Data.Length; i++)
{
Assert.AreEqual(cYdata[i], y.Data[i], delta);
}
//x.grad
Assert.AreEqual(cXgrad.Length, x.Grad.Length);
for (int i = 0; i < x.Grad.Length; i++)
{
Assert.AreEqual(cXgrad[i], x.Grad[i], delta);
}
//gamma.grad
Assert.AreEqual(cGammaGrad.Length, batchNormalization.Gamma.Grad.Length);
for (int i = 0; i < batchNormalization.Gamma.Grad.Length; i++)
{
Assert.AreEqual(cGammaGrad[i], batchNormalization.Gamma.Grad[i], delta);
}
//beta.grad
Assert.AreEqual(cBetaGrad.Length, batchNormalization.Beta.Grad.Length);
for (int i = 0; i < batchNormalization.Beta.Grad.Length; i++)
{
Assert.AreEqual(cBetaGrad[i], batchNormalization.Beta.Grad[i], delta);
}
}